AI Accessibility Scan Verification Lab

Automated accessibility tools can find useful problems, but a passing scan does not prove that a page is accessible. This lab asks students to use an AI-assisted scan, verify one finding independently and document the limits of the tool.

Goal

Use an accessibility scanner as evidence—not as an answer—and connect one reported issue to the page’s HTML and a relevant WCAG principle.

Activity

1. Choose a small public practice page or a local HTML file containing no private or student information. Save an untouched copy.
2. Start Stark’s no-card two-week trial and add the Stark connector from Claude’s connector directory.
3. Ask Claude and Stark to scan one URL or source file. Have Claude group the findings by severity and explain the highest-priority item in plain language.
4. Independently check one finding. Use the HTML inspector plus a manual test such as keyboard-only navigation, heading-outline review, form-label inspection or alternative-text review.
5. Correct one confirmed problem in the HTML. Scan again and compare the result.
6. Identify one thing the automated scan could not establish, such as whether alternative text communicates the image’s purpose or whether the keyboard order makes sense to a person.

Deliverable

Submit:

• the original and revised HTML fragment;
• the scanner’s finding and the applicable accessibility principle;
• the result of the independent manual check;
• a 100–150 word note explaining whether the AI-assisted recommendation was correct, incomplete or misleading.

Do not submit credentials, private URLs, student records or a full AI conversation containing sensitive context.

Discussion and safety

• Why is “no detected violations” different from “accessible”?
• Who remains responsible when an AI tool proposes a code change?
• What evidence should accompany an accessibility claim?
• Set a reminder to cancel or evaluate the Stark trial before it ends. The trial is not a permanently free plan.

Source material

This lab was first developed from the PTIR Morning Briefing for August 22, 2026. Stark’s announcement introduced a Claude connector that can scan URLs, source code, Figma files, mobile builds and Storybook libraries, then surface and discuss violations inside Claude. That workflow generated the lab because beginning web students need practice separating automated findings from independently verified accessibility evidence.

Consult the original announcement: Stark brings accessibility to Claude.